Stable EEG-Based biometric system using functional connectivity based on Time-Frequency features with optimal channels. (August 2022)

• Record Type: Journal Article
• Authors: Ashenaei, Roghaieh; Asghar Beheshti, Ali; Yousefi Rezaii, Tohid
• Is Part Of: Biomedical signal processing and control. Volume 77(2022)
• Publication Date: 2022-08
• Subjects: EEG biometric system -- Functional connectivity -- Wavelet coherence -- Partial Coherence -- Convolutional neural network -- Genetic algorithm; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.103790
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Bivariate features using functional connectivity is used to extract discriminative features. Proposed more realistic experimental paradigm. Performance is improved in terms of identification accuracy and stability issue. Genetic algorithm is used for channel reduction and the performance is not much degraded. Abstract: EEG-based biometric systems have received much attention during the last decades. Despite the positive results, EEG based biometric systems still have been not used in practice. Since, most of the existing studies use resting state signals or signals from the average of repeated trials, which limit their use in practice. Moreover, often univariate features which have limited discriminatory power are used in EEG based biometric systems. So, there has been a growing interest to extract distinct bivariate features from human brain areas. In this paper, due to the non-stationarity of EEG signals, we exploited time–frequency metrics for brain connectivity matrix to extract more discriminative features. In addition, to investigate the permanence and stability issues, we proposed a more realistic experimental paradigm in which signals of training and testing are recorded in two separate sessions and different states. Epochs have no overlap with each other in training and testing stages and accuracies were obtained from single trials whilst a multitude of published reports relied on average of repeated trials which are time-consuming. Two databases (self-recorded and public PhysioNet BCI) were used in this work. We compare our method with state-of-the-art methods and experimental results demonstrate the recognition rate above 99.50% which confirm the effectiveness of our approach. In this paper, we also exploited the genetic algorithm to select the minimum number of electrodes and despite of the reduction of EEG channels, identification performance of our proposed biometric system is degraded just 1%–2%.